Document 10463617

advertisement
Internat. J. Math. & Math. Sci.
Vol. 24, No. 4 (2000) 283–288
S0161171200003136
© Hindawi Publishing Corp.
DYNAMICS OF A CERTAIN SEQUENCE OF POWERS
ROMAN SZNAJDER and KANCHAN BASNYAT
(Received 14 September 1998 and in revised form 15 January 1999)
Abstract. For any nonzero complex number z we define a sequence a1 (z) = z, a2 (z) =
za1 (z) , . . . , an+1 (z) = zan (z) , n ∈ N. We attempt to describe the set of these z for which
the sequence {an (z)} is convergent. While it is almost impossible to characterize this
convergence set in the complex plane Ꮿ, we achieved it for positive reals. We also discussed
some connection to the Euler’s functional equation.
Keywords and phrases. Power sequence, dynamics, convergence.
2000 Mathematics Subject Classification. Primary 26A18, 39B12; Secondary 11B37, 37A45.
1. Introduction. In this paper, for any z ∈ Ꮿ\{0}, we consider the sequence
a1 (z) = z,
a2 (z) = zz = za1 (z) , . . . , an+1 (z) = zan (z),
n = 1, 2, . . . .
(1.1)
We wish to characterize the convergence set L := {z ∈ Ꮿ\{0} : {an (z)} is convergent}.
Since 1 ∈ L, this set is nonempty. For a complex value of z, the mapping z → an (z) is a
multivalued operation and an (z) usually represents an infinite set. If we agree to use
the main branch of the complex logarithm (as it is done, e.g., in MATLAB), the above
exponentiation becomes a single-valued operation. For arbitrary z, it is very easy to
run out of accuracy of any available computational software in a few iterations. Thus,
it is practically impossible to describe the convergence set L. Also, one can show that
the sequence {an (z)} cannot be obtained as a functional iteration, i.e., there exists
no function f : Ꮿ → Ꮿ such that an (z) = f [n] (z) = f ◦ f ◦ · · · ◦ f (z) (n-times). This
sequence has nasty numerical features. Nevertheless, the situation is not hopeless
when it comes to considering real values of an (z).
2. Main result. Our main goal is to demonstrate the following.
Theorem 2.1. We have L R+ = [e−e , e1/e ].
Before we proceed with the proof of Theorem 2.1, the following lemma is in order.
Lemma 2.2. For the function f (t) = ln(t)/t defined on R+ = (0, +∞), the following
hold true:
(a) range (f ) = (−∞, 1/e],
(b) for any 0 < c < 1/e, the equation f (t) = c has exactly two solutions, while for
c = 1/e exactly one, namely t ∗ = e.
Proof. The first derivative of the function f (t) is f (t) = (1 − ln(t))/t 2 , so that
f (t) = 0 if and only if t = e, f is increasing on (0, e] and decreasing on [e, +∞). Since
284
R. SZNAJDER AND K. BASNYAT
limt→0+ f (t) = −∞ and limt→+∞ f (t) = 0, we obtain statement (a). Since t ∗ = e is the
only maximizer of f (t) in (0, +∞), (b) follows from the intermediate value theorem
and the fact that limt→+∞ f (t) = 0. The proof is complete.
f (t)
0.4
1/e
0.3
0.2
0.1
0
(1) ∗
t
1 tc
(2)
=e
tc
t
−0.1
Figure 2.1.
(1)
(2)
Note that if tc and tc denote two distinct solutions of the equation f (t) = c,
(1)
(2)
then tc → 1+ and tc → +∞ as c → 0+ . This remark follows from the piecewise
monotonicity of the function f (t) and the fact that limt→1 f (t) = 0 and limt→+∞ f (t) =
0. By putting c = ln(x) in Lemma 2.2, we get the following.
Corollary 2.3. For any given x ∈ (1, e1/e ) there exist exactly two solutions of the
equation x α = α, while for x = e1/e , exactly one.
t
Lemma 2.4. Let K(x) := {0 ≤ t ≤ 1 : hx (t) = 0}, where hx (t) = x x − t and x > 0.
The following statements hold true:
(a) For x ∈ (0, e−e ), K(x) consists of three distinct elements.
(b) For x ∈ [e−e , 1), K(x) is a singleton.
Proof. Instead of giving a rigorous analytic proof similar to Bomberger [1], we
rather illustrate our point with an aid of appropriate graphs.
Case 1 (x ∈ (0, e−e )). See Figure 2.2.
Case 2 (x = e−e ). See Figure 2.3.
t
In fact, hx (t) = x t x x (ln(x))2 − 1, so he−e (1/e) = 0. Also, he−e (1/e) = 0.
Case 3 (x > e−e ). See Figure 2.4.
This completes our proof.
Proof of Theorem 2.1
Case 1 (0 < x < e−e ). By Lemma 2.4 (see Figure 2.2), there exist numbers 0 < α1 <
285
DYNAMICS OF A CERTAIN SEQUENCE OF POWERS
α2 < α3 < 1 such that
xx
αi
= αi
for i = 1, 2, 3.
(2.1)
While α2 is the fixed point of the function gx (t) = x t , α1 and α3 are points of period
2 (x α1 = α3 and x α3 = α1 ).
Generating the sequence {an (x)} by putting a1 (x) := x = gx (1) and an+1 (x) =
gx (an (x)) for n ≥ 1, as illustrated by Figure 2.5, we get
a1 (x) < a3 (x) < · · · < a2n−1 (x) α1 ,
a2 (x) > a4 (x) > · · · > a2n (x) α3 . (2.2)
Thus, {an (x)} diverges but consists of two complementary convergent subsequences.
Case 2 (x = e−e ). In this case, {an (x)} is convergent and the limit is 1/e. Indeed,
the sequence {an (x)} consists of two complementary subsequences and each of them
is convergent to a point of period 2. But in this case, α1 = α2 = α3 = 1/e, and the
two points of period 2 collapse to the fixed point of the function ge−e (t). Hence,
{an (x)} converges to 1/e. However its convergence is very slow (e.g., a1,000,000 (e−e ) ≈
0.36697888108297 and |a1,000,000 (e−e ) − 1/e| ≈ 0.00180).
Case 3 (e−e < x ≤ 1). For x < 1 we reason the same way as in Case 2, so the sequence {an (x)} converges to a unique fixed point of the function gx (t). For x = 1,
the statement is trivial.
Case 4 (1 < x < e1/e ). If {an (x)} is convergent to a limit α, {an+1 (x)} also converges to α, moreover, an+1 (x) = x an (x) converges to x α . Consequently, x α = α. By
Corollary 2.3, there are exactly two fixed points α1 and α2 (α1 < α2 ) of the function
gx (t). Obviously, 1 < α1 < e < α2 . Since gx (t) is increasing, 1 < x < α1 , and we check
by the math induction that a1 (x) < a2 (x) < · · · < an (x) < a1 , so the limit of the
sequence {an (x)} exists and is just a fixed point α1 .
0.04
x
0.02
hx (t)
0
α1
α2
α3
−0.02
−0.04
−0.06
Figure 2.2.
t
286
R. SZNAJDER AND K. BASNYAT
0.1
e−e
0.05
hx (t)
0
1/e
t
−0.05
−0.1
Figure 2.3.
Case 5 (x = e1/e ). As before, 1 < x < e and {an (x)} is increasing and bounded by
e, thus convergent to the only fixed point of ge1/e (t), namely e itself.
Case 6 (x > e1/e ). By Corollary 2.3, no fixed point of gx (t) exists, so the sequence
{an (x)} does not converge. Now, the proof is complete.
We note that in Case 4 of the above proof, α1 is an attracting fixed point, while
α2 is repelling fixed point for gx (t). In the limiting Case 5, the point α = e is left
0.1
x
0.05
hx (t)
0
t
−0.05
−0.1
−0.15
Figure 2.4.
287
DYNAMICS OF A CERTAIN SEQUENCE OF POWERS
attracting and right repelling. Our Corollary 2.3 and Lemma 2.4 have some interesting
relations to certain functional equations, one of them considered already by Euler (see
Sierpiński [2]):
(a) 0 < x < e−e . As we have already noticed, there exist exactly three zeros of the
α
function hx (t), say, α1 < α2 < α3 , so that x x i = αi for i = 1, 2, 3, see (2.1); α1
and α3 are points of period 2 of the function gx (t), i.e., x α1 = α3 and x α3 = α1 ;
(α2 is just the fixed point of the function gx (t)). Thus,
α
α
α
α
α 1 1 = x α3 1 = x α1 3 = α3 3 .
(2.3)
In other words, α1 and α3 satisfy the functional equation uu = v v . For a detailed
account on solving this equation we refer to Bomberger [1]. Observe that all αi ’s are
in (0, 1).
(b) 1 < x < e1/e . As we noted after Lemma 2.2, there exist two distinct fixed points
α1 and α2 of gx (t), and when x → 1+ , α1 → 1+ , and α2 → +∞. Since x α1 = α1 and
x α2 = α2 , we get
α
α
α
α
α1 2 = x α 1 2 = x α 2 1 = α 2 1 ,
(2.4)
thus, α1 and α2 are related by the equation uv = v u (u, v > 0), called the Euler
equation.
(c) e−e ≤ x ≤ 1. Now, α1 and α3 considered in (a) collapse to the same point and
both types of equations considered in (a) and (b) are automatically satisfied.
1
y
0.9
0.8
0.7
0.6
gx (t)
y =t
0.5
0.4
0.3
0.2
0.1
x
0
α1
α2
α3
Figure 2.5.
1 t
288
R. SZNAJDER AND K. BASNYAT
For the sake of completeness of this exposition we describe the set of solutions of
the Euler equation.
Proposition 2.5. The set of solutions of the Euler equation uv = v u , where u ≠
v, u, v > 0 is given by a parametrized family {(u, v) = α1/(α−1) , αα/(α−1) : α > 1}.
Proof. We note that if 0 < u < v ≤ 1, then u < uv < uu < v u , and a pair (u, v)
cannot be a solution. Similarly, when 0 < u ≤ 1 < v. Observe that ln(u)/u = ln(v)/v,
and if this common value is less than 1/e, by Lemma 2.2 there exist exactly two solutions u and v of this equation and 1 < u < e < v. We postulate a solution in the form
u = t and v = t α , where t > 1 and α > 0 are parameters. Thus,
α
t t = (t α )t = t αt ,
(2.5)
so t α = αt and, in consequence, t = α1/(α−1) . Hence, (u, v) = (α1/(α−1) , αα/(α−1) ). For
any 1 ≠ α > 0, α1/(α−1) > 1, so in order to keep u < v, α must be greater than 1. We
check that the pair (u, v) actually solves the equation:
αα/(α−1)
α1/(α−1)
α/(α−1)
1+1/(α−1)
uv = α1/(α−1)
= α1/(α−1)α
= α1/(α−1)α
= αα/(α−1)
= v u.
(2.6)
Now, we show that this parametrized family exhausts all possible solutions of the
Euler equation such that u < v. In fact, if α → 1+ , α1/(α−1) → e, and αα/(α−1) → e; if α →
+∞, α1/(α−1) → 1, and αα/(α−1) → +∞. The result now follows from the intermediate
value theorem.
Acknowledgements. We are very grateful to Dr. Denny Gulick from the University of Maryland, College Park and an anonymous referee whose suggestions and
comments were invaluable for improving the presentation of this paper. This research
was supported by the Fall 1997 MIE Faculty Grant.
References
[1]
[2]
J. D. Bomberger, On the solutions of aa = bb , Pi Mu Epsilon J. 9 (1993), no. 9, 571–572.
Zbl 875.26013.
W. Sierpiński, Elementary Theory of Numbers, 2nd ed., Elsevier Science Publ. Co., New
York, 1973.
Roman Sznajder: Department of Mathematics, Bowie State University, Bowie, Maryland 20715, USA
E-mail address: rsznajde@bowiestate.edu
Kanchan Basnyat: Department of Computer Science, Bowie State University, Bowie,
Maryland 20715, USA
E-mail address: kbasnyat@cs.bowiestate.edu
Download